English(EN) DP-Muon: Differentially Private Optimization via Matrix-Orthogonalized Momentum

新的DP-Muon方法增强了差分隐私优化

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-11 04:00

研究人员开发了一种新颖的差分隐私优化方法DP-Muon，该方法利用矩阵正交动量。该方法解决了将新高斯噪声应用于非线性矩阵变换时引入的均值失真问题。DP-Muon旨在减少这种偏差，在GPT-2上的实验表明，在各种隐私设置下，与Adam基线相比，在测试负对数似然方面取得了有利的结果。 AI

影响这项研究可能为大型语言模型带来更强大、更注重隐私的训练方法。

排序理由该集群包含一篇详细介绍差分隐私优化新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 11 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍差分隐私优化新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Same-day; Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

arXiv cs.LG TIER_1 English(EN) · Jihwan Kim, Chenglin Fan · 2026-09-11 04:00

DP-Muon：通过矩阵正交动量实现差分隐私优化

arXiv:2605.12994v2 Announce Type: replace Abstract: We study differentially private optimization with matrix-orthogonalized momentum. DP-Muon uses conventional global per-example clipping and one Gaussian gradient release per step; matrix updates and auxiliary updates are post-pr…